WO2017156627A1 - Automated method for assessing cancer risk using tissue samples, and system therefor - Google Patents

Abstract

An automated method and system for determining the risk of developing a cancer in a subject, the method comprising preparing a tissue sample obtained from the subject for visually identifying at least one biological marker associated with the cancer, digitally scanning the prepared tissue sample, analyzing the scanned image of the tissue sample to identify regions of interest, quantifying at least one parameter associated with the marker, and executing an algorithm using the quantified parameter to calculate a risk score, wherein the risk score is representative of the risk of the individual developing the cancer.

Description

AUTOMATED METHOD FOR ASSESSING CANCER RISK USING TISSUE SAMPLES, AND SYSTEM THEREFOR CROSS REFERENCE TO PRIOR APPLICATIONS [0001] The present application claims priority under the Paris Convention from US Application Number 62/308,182, filed March 14, 2016, the entire contents of which are incorporated herein by reference. FIELD OF THE DESCRIPTION [0002] The present description relates generally to methods of assessing tissue samples for the risk of developing cancer. More particularly, the description relates to computer- based and/or automated methods for conducting cancer risk assessments. BACKGROUND [0003] Despite being a low-prevalence cancer, the 5-year mortality rate of oral cancer remains around 40% in the US and around 50% in the world. The main reason for this high mortality rate is that oral cancers are typically not diagnosed until the cancer reaches advanced stages. Most patients (67-77%) do not seek medical attention until exhibiting symptoms of persistent pain associated with advanced stage cancer. Late stage cancers are very costly and difficult to treat. In the US, the cost for the first year treatment of oral cancer is on average $79k. Diagnosis at early stages can significantly lower the mortality rate and reduce treatment costs [0004] According to the current model of oral cancer development, the cancer is believed to go through a step-wise transformation from normal to pre-malignant to invasive carcinoma (Kalu et al., 2012). However, multiple pathways may also be involved, depending on the individual, the carcinogen and possibly other factors. The fact that most early stage cancers and pre-cancers are asymptomatic makes detection of such conditions difficult for clinicians. On the other hand, oral pre-malignant lesions (or oral potentially malignant lesions) are quite common, occurring in around 2.5% of the population (Weir, et al. 1987; McCullough, et al., 2010). It would be ideal if clinicians are able to identify cancer-prone lesions before these asymptomatic lesions become cancer. To further complicate things, the progression rate (i.e. the rate of lesions becoming cancers) is very low, below 5% per year (Hsue et al., 2007; Mehanna et al., 2009). The high mortality rate of oral cancer and the low transformation rate of pre-malignant lesions create a strong demand for a method or system that can reliably detect cancer-prone lesions. Unfortunately, till now such a reliable diagnostic method or system has not been available. [0005] Currently, there is no prognostic tool for oral pre-malignant lesions other than histopathological assessment which is in itself problematic. First, the predictive value of histopathological grading of pre-malignant lesions is very low. Generally, dysplastic lesions of higher grades have higher progression probability but have wide ranges. "A maximum of 50% of severe dysplasias, 30% of moderate dysplasias and very few (<5%) mild dysplasia are thought to progress to cancer" (Speight 2007). Thus, such grading is insufficient to provide clinicians with clear guidance on how to treat each individual pre-malignant oral lesion. Second, there is great inter- and intra-observer variation on grading.

Histopathological assessment by pathologists requires specialty experience, namely, oral pathology training. Pathologists experienced in grading other tissue biopsies are not necessarily competent in grading oral lesions. Variations in an individual's experience, combined with a lack of consensus on the features to complete the evaluation, has contributed to great inter- and intra-observer variation (Warnakulasuriya et al. 2008). As a result of this subjectivity, the same lesion may receive different diagnoses or treatments depending on which clinician assesses the lesion. The end result is that in clinics, clinicians have considerably difficulty in applying the grading to treatment options. Thus, in view of the low progression rate of a lesions, clinicians typically elect simply to monitor the lesion instead of taking pre-emptive measures. Consequently, most high-risk pre-malignant lesions are not diagnosed, treated or otherwise addresses, resulting in at least some of these lesions eventually transforming to cancer. [0006] Considerable efforts have been made to develop clinical and/or biomarker tests to enhance the prognostic range and to identify high-risk lesions. There is general consensus that high histopathological grading and advanced age are associated with high risk of progression to cancer (Schaaij-Visser, et al. 2010; Amagasa et al. 201 1 ;

Warnakulasuriya et al., 201 1). In some cases, controversies regarding gender have arisen (Amagasa et al. 201 1 ; Warnakulasuriya et al., 2015). A number of biomarkers have been proposed for association with oral lesion risk, including but not limited to hypermethylation of endothelin receptor type B (EDNRB) and kinesin family member 1 A (KIF1 A) (Pattani et al., 2010), loss of heterozygosity (Zhang et al., 2012), p16 methylation/HPV (Liu et al., 2015), DNA content (Xiao et al., 2015), and S100A7 (Kaur, R. et al. , 2014). However, none of these markers has yet been used commercially. Moreover, there is no quantitative method to predict the progression rate in individual dysplasia. [0007] US 2014/0235487 provides a quantitative method for determining the risk of developing oral cancers wherein certain morphological data of individual cells are used to calculate a risk score. [0008] There exists a need for a method, and system, that can provide a reliable means of determining the risk of a subject developing cancer, such as head and neck or oral cancer. SUMMARY OF THE DESCRIPTION [0009] In general, described herein is a method of automatically quantifying a risk score representative of the risk that a subject, or individual, will develop a cancer. The method is based on a tissue sample obtained from the subject and prepared to visually identify at least one biological marker. The preparation may involve, for example, staining. The prepared sample is then analyzed by a visualization system and the results of such system are processed according to an algorithm, which calculates the risk score. [0010] In one aspect, there is provided a method of prognosing the risk of developing a cancer in a subject, the method comprising: [0011] - preparing a tissue sample obtained from the subject for visually identifying at least one biological marker associated with the cancer; [0012] - digitally scanning the prepared tissue sample with a digital scanner to generate a scanned image of the sample; [0013] - analyzing the scanned image with an image analyzer to identify at least one region of interest and to quantify at least one parameter that characterizes the presence of the at least one biological marker; [0014] - transmitting the at least one quantified parameter to a processor, the processor being programmed to execute an algorithm for determining a risk score representative of the risk of the individual developing the cancer based on the at least one quantified parameter; and, [0015] - executing the algorithm to generate the risk score. [0016] In another aspect, there is provided a system for prognosing the risk of developing a cancer in a subject, the system comprising: [0017] - a scanning module for digitally scanning a biological sample obtained from the subject and generating a scanned image of the sample, the sample being pre-treated to visually identify at least one biological marker associated with the cancer; [0018] - a visualization module for analyzing the scanned image to identify at least one region of interest and to quantify at least one parameter that characterizes the presence of the at least one biological marker; and, [0019] - a processor programmed to execute an algorithm for determining a risk score representative of the risk of the individual developing the cancer based on the at least quantified parameter. BRIEF DESCRIPTION OF THE FIGURES [0020] The features of certain embodiments will become more apparent in the following detailed description in which reference is made to the appended figures wherein: [0021] Figure 1 illustrates the automated region of interest (ROI) selection. [0022] Figure 2 illustrates the automated labelling of the ROI. [0023] Figure 3 illustrates an example of overlapping ROIs. [0024] Figures 4a and 4b illustrate examples of input screens from the Visiopharm™ software used for the image analysis in the example. [0025] Figures 5a and 5b shows the Kaplan-Meier survival curves for the two groups of cases studied in the example. DETAILED DESCRIPTION [0026] The terms "comprise", "comprises", "comprised" or "comprising" may be used in the present description. As used herein (including the specification and/or the claims), these terms are to be interpreted as specifying the presence of the stated features, integers, steps or components, but not as precluding the presence of one or more other feature, integer, step, component or a group thereof as would be apparent to persons having ordinary skill in the relevant art. [0027] As discussed above, oral pre-malignant lesions (OPLs) are quite common, frequently asymptomatic, and generally detected during routine oral exams. While the transformation rate of pre-cancerous lesions to cancerous lesions is less than 5% per year, most early stage cancers and OPLs are also asymptomatic, making identification more difficult. Most OPLs do not require aggressive treatment; however, preventing the transformation to malignancy is key to impacting oral cancer morbidity and mortality.

Furthermore, the high mortality rate associated with oral cancer and the low transformation rate of OPLs creates a need for reliable assessments that more accurately identify lesions at high-risk of transformation, separating these lesions from those at lesser transformation risk. The standard of care for OPL risk assessment, namely, dysplasia grading by histopathology, is subject to intra- and inter-observer variation as well as significant overlap between grades, thereby rendering such known method unreliable as a prognostic tool. The method and system described herein has been developed as a prognostic tool to meet these needs. [0028] The present description relates to a method, in particular an automated method, of determining a subject's (i.e. patient's) risk of developing a cancer. In one aspect, the description involves conducting an automated image analysis of a tissue sample that has been stained for one or more biological markers associated with the cancer. Such markers may be cell morphologic properties (such as protein or nucleic acid markers for example. Preferably, the one or more biological markers are one or more protein biomarkers. The cancer in question may be any cancer to which the biological marker has been correlated. The correlation may be a reduction or increase in the amount or concentration of the marker when the cancer is present. [0029] As mentioned above, a preferred biological marker is one or more protein markers. In such case, the image analysis would involve treating a biological sample obtained from the subject with an agent for visually detecting the presence or absence of the protein marker. In one example, the means of visually detecting the presence of a given protein involves the use of 3,3'-diaminobenzidine (or "DAB"). In such case, the sample obtained from the subject is contacted with an antibody specific to at least one of the protein markers under investigation, the antibody being conjugated with a peroxidase enzyme. Once the antibody is allowed to bind to the protein(s) of interest, the sample is treated with hydrogen peroxide and DAB. The DAB is thereby oxidized, forming a brown precipitate. The brown precipitate can then be visually detected and the presence and intensity of such color enables the detection of the protein(s). It will be understood that this specific visualization technique is only one example of the types of marker detection methodologies that would be known to persons skilled in the art. [0030] According to the described method, once a particular visualization technique has been chosen and the necessary reagents deployed on the patient sample, the stained sample is then analyzed by an automated system. One such analytical visualization system is commercially available from Visiopharm™, which is discussed further below. However, it will be understood that the present description is not limited to any particular visual detection system, apparatus or method. Indeed, any system, apparatus or method that is capable of detecting stained samples and generating analytical data may be used in accordance with the present description. [0031] The visualization system first identifies a region of interest (ROI) within the sample. The ROI is preferably identified or delineated based on the detected concentration of the marker(s) in question in the tissue as well as the location of the marker(s) in the tissue. Such location information would preferably also include topologic location information concerning the location of the marker(s) within the tissue. For example, the information may indicate the amount of the marker(s) in the epithelium This detection may involve a "heat map" methodology, wherein the intensity of a stain color (e.g. brown, in the case of DAB) indicates concentration of the marker. The ROI determination would preferably require a threshold intensity to be determined initially. Once the threshold intensity is set, the visualization system would automatically identify the ROI. Such ROI may be visually represented by one or more boundary lines to identify areas of marker concentration that exceed the concentration threshold. These regions, or "hot spots", superimposed on a digital image of the sample. [0032] After this, the visualization system then conducts a further data acquisition step on the ROIs that are identified. In this step, the system calculates the value of at least one parameter representative of the presence of the biological marker. In one aspect, the visualization system calculates the values of at least two parameters, namely: (1) a first parameter, P1 , comprising for example a value representing the percentage of the ROI area that is positive for the marker(s) in question; and (2) a second parameter, P2, comprising for example a value representing the average cell size of the cells within the ROI. As will be understood, this step can be conducted automatically using software associated with the visualization system. It will be understood that in other embodiments, further parameters may be determined either quantitatively or qualitatively. For example, in one embodiment, the presence of the marker(s) in question may be determined at the subcellular level, thus providing data on the sub-cellular localization of the marker(s) within the cells in the ROI. It will be understood that the description is not limited to the number of other parameters that may be incorporated into the analysis. [0033] Once the values of at least the two parameters, P1 and P2, are calculated, the visualization system transmits such data to a processor for further processing. The processor may be part of or otherwise associated with the same hardware system used to conduct the visualization procedure or it may be associated with a separate hardware component such as a local or remote computer or server. The processor is programmed to receive the values of P1 and P2 to perform one or more further mathematical operations on same. In particular, the processor of the present description is encoded to execute a first algorithm to calculate a risk score, RS. The risk score is calculated by multiplying each of the parameters, P1 and P2, with a suitable weighting factor and then subtracting the weighted average cell size value from the weighted marker-positive area value. In other words, the following function is executed: RS = (W! X P1) - (w₂ x P2) wherein, w₁ is the weighting factor (or weight) for P1 and w₂ is the weighting factor for P2. It will be understood that the weighting factors can vary based on the marker or markers in question and on the desired sensitivity of the analysis. The values for the weighting factors can be calculated using any statistical modelling techniques or methods as would be known in the art. For example, the calculation of the weighting factors may involve linear regression or Cox regression methods based on a given data set. In the Example contained herein, weighting factors (i.e. w₁ and w₂) were calculated in the context of assessing the risk of developing oral cancer based on the protein S100A7 as the cancer marker. These weighing factors were calculated using a multivariate Cox regression model based on the test samples mentioned in the Example. The description is not limited to any particular method or methods for calculating weighting factors. [0034] The processor is also programmed to calculate a probability value, PV, that the cancer will develop within a time period t. The equations used to calculate PV are as follows: PV = 1 - S(t)

5(t) = exp(-H(t))

[0035] In the above equations:

- PV is, as indicated above, the probability of developing cancer within a time period t.

- S(t) is the probability of not developing cancer within the time period t.

- H(t) is the cumulative baseline hazard up to time t. [0036] Further description of these equations is provided below in relation to an example for illustration purposes. [0037] As described above, the system for performing the visualization of the sample need not necessarily be physically located together with the processor that conducts the aforementioned risk calculations. Thus, in one aspect, the visualization procedure can be performed at a lab located in one location. The results, i.e. quantified parameters P1 and P2, can then be transferred or transmitted to another location where the processor may be situated, which processor can the execute the subsequent calculations. In this way, the processor calculating the risk value(s) may be located a separate office, such as an office of a data analysis service provider. The service provider can then provide a clinician or physician or patient with the calculated risk value. The above description has involved two entities, the lab and the service provider. However, it will be understood that any number (i.e. one or more entities) may be involved in the data analysis/manipulation process. [0038] As described in the example below, the above automated method was used to determine the risk of developing oral cancer in patients, wherein the protein marker S100A7 was utilized. This marker has been described as a known biomarker for head and neck cancer (Ralhan et al., 2008; Tripathi et al., 2010) and later for oral mucosal dysplasia (Kaur et al. 2014). It is believed that the method presented herein is the first to quantitatively determine a risk for developing oral cancer in oral pre-malignant lesions using the marker S100A7. In addition, in view of the significant findings from the use of the present method, it is believed that it can be used in determining the development risk, in other words prognosis, of other cancers. It will be understood that the choice of biological markers will depend on the cancer under review. [0039] As will be understood, each of the steps of preparing the tissue sample, scanning the tissue sample, analyzing the scanned image and processing the quantified parameter(s) can be performed at discrete locations or in the same location as needed. Further any group or subset of the steps can be performed at the same location or at different locations. In either case, it will be understood that the data generated by each step can be transmitted in any manner as would be commonly known. For example, the data from one step can be passed to the other step over any data communication network, or may be physically transported from one location to another by means of a memory device, such as a USB device, disk etc. The present description is not limited by the means by which data is transmitted. [0040] As would be understood, the description also encompasses a system for performing the aforementioned method steps. Such a system would include a digital scanning system or device (also referred to herein as a scanning module, which would be understood to encompass hardware and associated software), that is capable of scanning a biological sample that has been treated to visually identify one or more biological markers. The system would also include an image analysis system, or visualization system (also referred to herein as a visualization module, which would be understood to encompass hardware and associated software), for analyzing the digital image generated by the scanning system. The visualization system may, for example, identify the regions of interest in the scanned image and also generate one or more quantified parameters representative of the presence and concentration etc. of the marker. The system described herein would preferably also include a processor for receiving the quantified parameter(s) and for executing an algorithm that calculates a risk score based on the value(s) of the

parameter(s). The risk score would be representative of the risk to the subject of developing the cancer. [0041] The present description will now be illustrated by means of the following example. It will be understood that the example is not meant to limit the scope of the description in any way. [0042] Example [0043] Summary [0044] Oral pre-malignant lesion (OPL) biopsy samples from 150 cases with a follow-up history of up to 12 years were used. Immunohistochemistry for the biomarker S100A7 on tissue biopsy slides and tissue microarrays were performed at Mount Sinai Hospital in Toronto, Ontario, Canada. The slides were then digitally scanned on a slide scanner and images were visualized and analyzed using Visiopharm™ VIS. [0045] All statistical analyses and model building were conducted using the R package (version 3). Stepwise Cox Regression was used to select the parameters. A multivariate Cox Regression model was fitted to selected parameters and the C-index was used to assess the model. Estimated Log Relative-Hazards from the Cox model were referred to as risk scores and used in the cut-off selection stage to classify all cases into three risk groups: low, intermediate, and high. The Nelson-Aalen-Breslow estimate, used to calculate the baseline cancer-free survival curve, was combined with the calculated risk score to produce the expected cancer-free survival probability for each case. The Aalen-Link-Tsiatis estimate, used to estimate the variance of expected cancer-free survival probability, provided the 95% confidence interval (CI) of the cancer-free survival curve. [0046] From the 150 cases, the 95% CI of mild, moderate, and severe dysplasia grades based on histopathological assessment, overlapped extensively throughout the first 60 months, indicating ineffective differentiation. In contrast, the 95% CIs of the groups classified according to the presently described method had minimal overlaps at month 60, achieving better differentiation. The performance of present method was evaluated by an internal validation study using the split-sample technique. Comparing the C-index (time-to-event response) and Area Under the Curve (AUC; binary response), the risk scores calculated by the present method were found to be more objective and discriminatory than

histopathological dysplasia grading. Furthermore, the present method was found to outperform histopathological dysplasia grading in two clinical indices. The sensitivity between the low-risk vs. non-low-risk using the present method was 96% compared to the mild vs. non-mild dysplasia grading which was 75%, with a negative predictive value of 80% and 59%, respectively. [0047] The present method is believed to better categorize a patient's 5-year risk of OPLs progressing to cancer. The method can be easily incorporated into clinical practice as no additional tissue samples are needed for the assessment. [0048] Materials and methods [0049] Tissue biopsy slides [0050] 150 samples were used in this application. The tissue biopsy slides, tissue microarrays and immunohistochemistry staining have been described previously (Kaur, Sawhney et al. 2013). The staining was performed at a commercial clinical lab accredited in the province of Ontario, Canada. The slides were then digitally scanned on a Hamamatsu Nanozoomer-XR™ slide scanner. The images of the slides were visualized using

Visiopharm™ VIS software (version 5.0.1 .1 122, Hoersholm, Denmark). Clinical information for each sample such as dysplasia grading, gender, age, etc. was provided by Mount Sinai Hospital. This project including the informed consent form was approved by Mount Sinai Hospital Research Ethics Board (project 13-0197-E). [0051] Visiopharm™ APPs for automated regions of interest identification, cell classification, and counting [0052] The ROI selection, cell classification, and counting were performed using Visiopharm VIS. Five independent Visiopharm™ APPs were used in tandem to carry out the process. [0053] APP1 : Whole Tissue selection. The tissue on a slide was outlined for further analysis. Glass with no tissue or with staining debris was excluded. [0054] APP2: Whole Tissue to DAB Area conversion. DAB (3,3'-diaminobenzidine) positive regions (intensity below a user-defined threshold) on the tissue were selected. [0055] APP3: DAB Area to cell classification. Cell nuclei were labeled in the DAB positive regions. [0056] APP4: DAB Heat mapping. A heat map was generated throughout the DAB positive region based on the density of nuclei (density was defined as number of nuclei per 10 μηι diameter circle). Five hottest spots (with highest density of nuclei) were selected and five 500 μηι diameter circles were created the centers at each of the hot spot. These circles might overlap, giving irregular shapes (Figures 1 and 3). [0057] APP5:Nuclei classification and Positive DAB% calculation. Only tissue found inside the regions of interest (ROIs) as determined in APP4 were included at this stage. Areas void of tissue (background) was designed to be absent from any analysis. The nuclei were re-labeled which not necessarily overlapped with the labels in APP3. Cytoplasm were classified and labeled as either positive or negative for DAB staining. Data on two parameters were recorded, the average size of cells and the percentage of DAB positive areas versus the total area of ROIs (Figure 2). [0058] Algorithm [0059] The present algorithm has two major steps. The first step, "image analysis", is used to obtain measurements of S100A7 from slide images. The second step, "risk calculation", is used to feed the measurements from the first step to a formula to produce cancer progression probability. These two steps are discussed further below. [0060] 1) Image analysis. In Visiopharm™, first "batch process" is selected. Five applications, or APPs, are added sequentially into the Batch process window and the APPs are executed. Figures 4a and 4b illustrate sample input screens for selecting the desired APPs. [0061] 2) Risk calculation. The calculation takes values DAB_Pos_Per and

Average_cell_size and inputs into the following formula to provide the cancer progression risk over the next five years: [0062] Risk Score = [2.9807 x (DAB_Pos_Per)] - [0.005636 x {Average _cell_size)] [0063] In the above formula, DAB_Pos_Per represents the percentage of DAB positive area within the region of interest (ROI) and Average_Cell_Size represents the average size of the cells (both DAB positives and negatives) within the ROI. [0064] 3) Cancer progression risk calculation. [0065] Probability of Cancer before time t = 1— S{t) [0066] Where S(t) is the probability of not having cancer before time t, given by the following formula: [0067] 5(t) = exp(-H(t)) [0068] Where H(t) is the cumulative baseline hazard up to time t, given by the following formula: [0069] H{t) = f* h₀(u) exp(2.9807 * OAB Pos Per - 0.005636 * Average _cell_size) d(u) [0070] In the above formula, h₀(t) is the baseline hazard, depends on the time point t. [0071] Statistical analysis and modeling [0072] All statistical analyses and model building were conducted using R package (version 3.2.2). Stepwise Cox regression was used to select parameters. A multivariate Cox regression model was fitted to the selected parameters and C-index was used to assess the model. Estimated log relative- hazards from the Cox model were referred as risk scores and were used in the cut-off selection stage to classify all cases into three risk groups. Statistical analysis was based on 150 cases and parameters based on automatically outlined ROIs. Cox-regression was used to develop risk score formula. [0073] Results

[0074] Table 1 summarises the results of the 150 cases that were reviewed in this study. [0075] Table 1

Total Dysplasia grading

Mild Moderate Severe n 150 65 54 31

Cancer (% of cancer cases) 60 (40.0%) 14 (21.5%) 24 (44.4%) 22 (71.0°

Gender Male 76 32 27 17

Female 74 33 27 14

Site Tongue 99 46 39 14

Others 51 19 15 17

Age (year) mean 59 59 60 60

median 60 60 62 58

range [32, 88] [33, 88] [37, 88] [32, 83]

DAB_Pos_Per mean 47 40 48 57

median 50 42 48 66

range [1 , 93] [1 , 87] [1 , 92] [2, 93]

Average cell size mean 163 183 147 150 median 150 164 143 136

range [75, 492] [87, 492] [75, 242] [79, 261 ]

[0076] Classification of dysplasia cases into three risk groups [0077] A common difficulty with dysplasia grading is that the three grades are not good predictors of cancer progression with extensive overlap between the mild, moderate and severe groups. With the risk scores developed, two cut-offs were selected to classify all cases into 3 risk groups: low risk, intermediate risk and high risk. [0078] 1 . For all patients: choose a high cut-off to differentiate high risk group and non- high (intermediate and low) risk group [0079] a) Specificity > 85% [0080] b) P value of Log Rank test between high risk group and non-high risk group < 0.05 [0081] c) Among cut-offs that meet the first two criteria, choose the one that gives the best balanced accuracy, defined as the average of sensitivity and specificity [0082] 2. For patients in non-high risk group: choose a low cut-off to differentiate intermediate risk group and low risk group [0083] a) Sensitivity> 90% [0084] b) P value of Log Rank test between intermediate risk group and low risk group < 0.05 [0085] c) Among cut-offs that meet the first two criteria, choose the one that gives the best balanced accuracy [0086] Table 2 summarizes the risk group cut-offs. [0087] Table 2

Cancer progression probability in 5 years Risk group probability≥ 60% High risk

19%≤ probability < 60% Intermediate risk

probability < 19% Low risk [0088] The algorithm used in the present example was found to classify dysplasia more accurately than histopathological grading in relation to cancer progression. [0089] Once the 5-year cancer progression probability is calculated, a patient can be classified into one of three risk groups. The Nelson-Aalen-Breslow estimate was used to calculate the baseline cancer free survival curve. Based on risk score and the baseline cancer free survival curve, the expected cancer free survival probability for a patient can be calculated. [0090] The Aalen-Link-Tsiatis estimate was used to estimate the variance of expected cancer free survival probability. Based on the variance and expected cancer free survival probability, the 95% confidence interval (CI) of the cancer free survival curve is generated. A log-log transformation is used when generating Cls so that the Cls are bound by [0%, 100%]. The 5-year expected cumulative cancer progression probability curve is then plotted as 100% - cancer free survival probability (%), with 95% CI calculated as [100% - 95% CI of cancer free survival probability upper bound, 100% - 95% CI of cancer free survival probability lower bound]. [0091] Using the 150 cases, the 95% Cls of risk categories based on histopathological assessment overlapped extensively throughout the first 60 months, indicating inefficient differentiation. In contrast, 95% Cls of the test groups had minimal overlapping at month 60, thus achieving better differentiation. Figures 5a and 5b illustrate the Kaplan-Meier survival curves for the two groups. [0092] Performance of Algorithm [0093] The performance of the algorithm was evaluated by an internal validation study using split-sample technique. In the validation study, the data set with 150 cases was randomly split into two data sets: a training set with 1 10 cases and a testing set with 40 cases. The algorithms (both Cox regression model and cut-off selection steps) were rebuilt from scratch using the training set. The cases in the testing set were classified using the rebuilt algorithm into three risk groups. This procedure was repeated 100 times. C-index (using time-to-event response) and AUC (using binary response) of the risk scores were recorded for each testing set. [0094] As shown in Table 3, the calculations of the C-index and AUC indicate that the risk scores based on the algorithm were more objective and discriminatory than the known dysplasia grading method. [0095] Table 3

Predictor C-index AUC

Dysplasia grading (mild, moderate, 0.67 0.67 severe)

Algorithm risk score 0.72 0.72 [0096] Clinical performance [0097] As illustrated in Tables 4 and 5, some of the improvements offered by the present algorithm-based method over the known dysplasia grading method are: [0098] 1) Higher sensitivity in the low cut-off, 92% (vs 74%) , a 25% improvement. [0099] 2) Reduced false negative rate, 8% (vs 26%), by more than two-thirds. [00100] 3) Reduced false omission rate 24% (vs 39%), by almost two-fold. [00101] Table 4: Dysplasia grading results

Dysplasia grading

Sensitivity Specificity PPV NPV

Severe vs Moderate + 37% 83% 76% 48%

Mild

Severe + Moderate vs 75% 53% 69% 59%

Mild [00102] Table 5: Results from algorithm

Straticyte

Sensitivity Specificity PPV NPV

High vs 42% 85% 80% 51%

Intermediate + Low

High + 96% 24% 64% 80%

Intermediate vs

Low

[00103] Although the above description includes reference to certain specific

embodiments, various modifications thereof will be apparent to those skilled in the art. Any examples provided herein are included solely for the purpose of illustration and are not intended to be limiting in any way. Any drawings provided herein are solely for the purpose of illustrating various aspects of the description and are not intended to be drawn to scale or to be limiting in any way. The scope of the claims appended hereto should not be limited by the preferred embodiments set forth in the above description, but should be given the broadest interpretation consistent with the present specification as a whole. The disclosures of all prior art recited herein are incorporated herein by reference in their entirety.

[00104] References

[00105] Amagasa, T., M. Yamashiro, et al. (201 1). "Oral premalignant lesions: from a clinical perspective." Int J Clin Oncol 16(1): 5-14.

[00106] Hsue, S. S., W. C. Wang, et al. (2007). "Malignant transformation in 1458 patients with potentially malignant oral mucosal disorders: a follow-up study based in a Taiwanese hospital." J Oral Pathol Med 36(1): 25-29.

[00107] Kalu U. E., Ogbureke, C. B. (2012). Overview of Oral Cancer. Oral Cancer. K. U. E. Ogbureke. Janeza Trdine 9, 51000 Rijeka, Croatia, InTech: 9.

[00108] Kaur, J., A. Matta, et al. (2014). "S100A7 overexpression is a predictive marker for high risk of malignant transformation in oral dysplasia." Int J Cancer 134(6): 1379-1388. [00109] Liu, H., X. W. Liu, et al. (2015). "P16 Methylation as an Early Predictor for Cancer Development From Oral Epithelial Dysplasia: A Double-blind Multicentre Prospective Study." EBioMedicine 2(5): 432-437. [00110] McCullough, M. J., G. Prasad, et al. (2010). "Oral mucosal malignancy and potentially malignant lesions: an update on the epidemiology, risk factors, diagnosis and management." Aust Dent J 55 Suppl 1 : 61 -65. [00111] Mehanna, H. M., T. Rattay, et al. (2009). "Treatment and follow-up of oral dysplasia - a systematic review and meta-analysis." Head Neck 31 (12): 1600-1609. [00112] Pattani, K. M., Z. Zhang, et al. (2010). "Endothelin receptor type B gene promoter hypermethylation in salivary rinses is independently associated with risk of oral cavity cancer and premalignancy." Cancer Prev Res (Phila) 3(9): 1093-1 103. [00113] Ralhan, R., L. V. Desouza, et al. (2008). "Discovery and verification of head-and- neck cancer biomarkers by differential protein expression analysis using iTRAQ labeling, multidimensional liquid chromatography, and tandem mass spectrometry." Mol Cell

Proteomics 7(6): 1 162-1 173. [00114] Schaaij-Visser, T. B. , J. F. Bremmer, et al. (2010). "Evaluation of cornulin, keratin 4, keratin 13 expression and grade of dysplasia for predicting malignant progression of oral leukoplakia." Oral Oncol 46(2): 123-127. [00115] Speight, P. M. (2007). "Update on oral epithelial dysplasia and progression to cancer." Head Neck Pathol 1 (1): 61 -66. [00116] Tripathi, S. C , A. Matta, et al. (2010). "Nuclear S100A7 is associated with poor prognosis in head and neck cancer." PLoS One 5(8): e1 1939. [00117] Warnakulasuriya, S., J. Reibel, et al. (2008). "Oral epithelial dysplasia classification systems: predictive value, utility, weaknesses and scope for improvement." J Oral Pathol Med 37(3): 127-133. [00118] Warnakulasuriya, S., T. Kovacevic, et al. (201 1). "Factors predicting malignant transformation in oral potentially malignant disorders among patients accrued over a 10-year period in South East England." J Oral Pathol Med 40(9): 677-683. [00119] Warnakulasuriya, S. and A. Ariyawardana (2015). "Malignant transformation of oral leukoplakia: a systematic review of observational studies." J Oral Pathol Med. [00120] Weir, J. C, W. D. Davenport, et al. (1987). "A diagnostic and epidemiologic survey of 15,783 oral lesions." J Am Dent Assoc 1 15(3): 439-442. [00121] Xiao, X., L. Shi, et al. (2015). "DNA content status using brush biopsy with image cytometry correlated with staging of oral leukoplakia: a preliminary study." Oral Oncol 51 (1): 59-63. [00122] Zhang, L, C. F. Poh, et al. (2012). "Loss of heterozygosity (LOH) profiles- validated risk predictors for progression to oral cancer." Cancer Prev Res (Phila) 5(9): 1081 - 1089.

Claims

WE CLAIM:

1 . A method of prognosing the risk of developing a cancer in a subject, the method comprising:

- preparing a tissue sample obtained from the subject for visually identifying at least one biological marker associated with the cancer;

- digitally scanning the prepared tissue sample with a digital scanner to generate a scanned image of the sample;

- analyzing the scanned image with an image analyzer to identify at least one region of interest and to quantify at least one parameter that characterizes the presence of the at least one biological marker;

- transmitting the at least one quantified parameter to a processor, the processor being programmed to execute an algorithm for determining a risk score representative of the risk of the individual developing the cancer based on the at least one quantified parameter; and,

- executing the algorithm to generate the risk score.

2. The method of claim 1 , wherein the step of analyzing is performed automatically following the step of scanning.

3. The method of claim 1 or 2, wherein the digital scanner, the image analyzer and the processor are connected together.

4. The method of claim 1 or 2, wherein the digital scanner, the image analyzer and the processor are connected together over a data communication network.

5. The method of any one of claims 1 to 4, wherein the at least one parameter represents: an area within the at least one region of interest comprising the at least one biological marker; an average size of the cells within the at least one region of interest; or a combination thereof.

6. The method of any one of claims 1 to 4, wherein the at least one parameter comprises two parameters.

7. The method of claim 6, wherein the two parameters represent (i) an area within the at least one region of interest comprising the at least one biological marker, and (ii) an average size of the cells within the at least one region of interest.

8. The method of any one of claims 1 to 7, wherein the cancer is oral cancer and the at least one biological marker comprises S100A7.

9. A system for prognosing the risk of developing a cancer in a subject, the system comprising:

- a scanning module for digitally scanning a biological sample obtained from the subject and generating a scanned image of the sample, the sample being pre-treated to visually identify at least one biological marker associated with the cancer;

- a visualization module for analyzing the scanned image to identify at least one region of interest and to quantify at least one parameter that characterizes the presence of the at least one biological marker; and,

- a processor programmed to execute an algorithm for determining a risk score representative of the risk of the individual developing the cancer based on the at least quantified parameter.

10. The system of claim 9, wherein execution of the visualization module is automatically initiated following execution of the scanning module.

1 1. The system of claim 9 or 10, wherein the scanning module, visualization module and processor are connected together.

12. The system of claim 9 or 10, wherein the scanning module, visualization module and processor are connected together over a data communication network.